🏷️ Fine-tuning NER using PyTorch

🎯 Overview

A Named Entity Recognition (NER) model fine-tuned using PyTorch and BERT on the CoNLL-2003 dataset. The project demonstrates how to train and evaluate a state-of-the-art NER model using modern deep learning techniques.

🛠️ Tech Stack

• Framework: PyTorch
• Model: google-bert/bert-base-cased
• Libraries:
  • transformers
  • seqeval
  • torch
• Hardware: NVIDIA 1650 Max-Q (4GB GPU)
• Environment: Jupyter Notebook + CUDA

📊 Dataset

Using the CoNLL-2003 dataset which includes:

• English Data: Reuters news stories (Aug 1996 - Aug 1997)

  • Training set: End of August 1996
  • Test set: December 1996
  • Raw data: September 1996

• German Data: Frankfurter Rundshau newspaper

  • All sets: End of August 1992
  • Raw data: September-December 1992

📈 Evaluation Metrics

Using seqeval for evaluation:

• F1 Score
• Recall Score
• Precision Score

Prerequisites

• Python 3.8+
• CUDA-capable GPU
• 4GB+ GPU Memory